IntHTE: Integrative analysis of the HTE (IntHTE)

In shuyang1987/IntegrativeHTE: Integrative analysis of the heterogeneous treatment effect (HTE)

Description

Implements integrative HTE analysis combing randomized trial and real-world data.

Usage

IntHTE(A, X, Y, delta, family.oc = "gaussian", gamma1 = 0.9,
  gamma2 = 2, nptb = 50, nboots = 50, alpha.CI = 0.05)

Arguments

A	is a vector of treatment ( (n+m) x 1).
X	is a matrix of covariates without intercept ( (n+m) x p).
Y	is a vector of outcome ( (n+m) x 1).
delta	is a vector of the binary indicator of belonging to the randomized trial (RT) sample; i.e., 1 if the unit belongs to the RT sample, and 0 otherwise ( (n+m) x 1).
family.oc	specifies the family for the outcome model. "gaussian": a linear regression model for the continuous outcome. "binomial": a logistic regression model for the binary outcome.
gamma1	is the parameter in the violation test such that T > chisq_1-gamma1 indicates the bias of the RW analysis is significant.
gamma2	is a smooth parameter in using the normal cummulative distribution with mean 0 and variance gamma2^2 function to approximate the indicator function.
nptb	is the number for replication-based method for estimating Sigma_SS.
nboots	is the number of bootstrap samples.
alpha.CI	is the level of percentile bootstrap confidence interval.

Details

Details will be provided in the reference paper.

Value

• est: the HTE estimators including i) the covariate adjustment estimator; ii) the RT estimating equation estimator; iii) the RT&RW estimating equation estimator; iv) the elastic integrative estimator.

• ve: variance estimate for est.

• boot.CI: 1-alpha.CI percentiel bootstrap confidence interval for the elastic integrative estimator.

References

A reference paper will come up soon.

Examples

library(MASS)
library(rootSolve)
library(stats)

set.seed(1234)

## population size
n <- 1e5
beta0<- c(0,1,1)
psi0 <- c(1,1,1)
family.oc <- "binomial"

X.pop <- cbind(rnorm(n,0,1),rnorm(n,0,1))
if( family.oc=="Gaussian" ){
 Y0   <-  cbind(1,X.pop)%*%beta0+rnorm(n,0,1)
 Y1   <-  cbind(1,X.pop)%*%beta0+cbind(1,X)%*%psi0+rnorm(n,0,1)
}
if( family.oc=="binomial" ){
 lY0   <- -cbind(1,X.pop)%*%psi0
 lY1   <-  cbind(1,X.pop)%*%psi0
 pY0 <- exp(lY0)/(1+exp(lY0) )
 pY1 <- exp(lY1)/(1+exp(lY1) )
 Y0    <- rbinom(n,1, pY0 )
 Y1    <- rbinom(n,1, pY1 )
}

## RCT data
eS <- expoit(-cbind(1,X.pop)%*%c(5.5,1,1))
mean(eS)*n
S <- sapply(eS,rbinom,n = 1, size = 1)
S.ind <- which(S==1)
n.t <- length(S.ind)
n.t
X.t <- X.pop[S.ind,]
Y0.t <- Y0[S.ind]
Y1.t <- Y1[S.ind]
A.t  <- rbinom(n.t,1,0.5) # randomly assign treatment to trial participant
Y.t  <-  Y1.t*A.t+Y0.t*(1-A.t)

## OS data
m <- 5000
P.ind <- sample(1:n,size = m)
X.os <- X.pop[P.ind,]
Y0.os <- Y0[P.ind]
Y1.os <- Y1[P.ind]

eA <- expoit( cbind(1,X.os)%*%c(1,-1,-1))
mean(eA)

A.os <- sapply(eA,rbinom,n=1,size=1)
Y.os <- Y1.os*A.os+Y0.os*(1-A.os)

A<-c(A.t,A.os)
X<-rbind(X.t,X.os)
Y<-c(Y.t,Y.os)
delta<-c(rep(1,n.t),rep(0,m))

gamma1=0.9
gamma2=2
nptb=5
nboots=2
alpha.CI=0.05

IntHTE(A, X, Y, delta, family.oc,
gamma1, gamma2, nptb, nboots, alpha.CI)

shuyang1987/IntegrativeHTE documentation built on April 13, 2020, 1:43 a.m.